Biological Resources Objectives:
The primary objective of the Biological Resource Program is to understand the cause/effect relations between the operation of the Glen Canyon dam and the downstream aquatic and terrestrial ecosystem, and to develop a model that can predict ecological effects for different dam operations. This objective is approached in three ways: (1) inventory of biologic resources and, together with related physical resource data, development of a conceptual model that links biotic and abiotic components; (2) research on and development of cause/effect relations between dam operations and the ecology and testing the validity of the observed relations under various dam operations; and (3) monitoring both long- and short-term ecosystem behavior to determine if models are predictive for both natural (tributary) and dam perturbations.
Parameters Measured and Methods Employed:
The parameters that are measured are those deemed to be stressors, directly or indirectly, that have been previously established or suspected of having cause and effect relations, such as sediment load, temperature, flow discharge, inter-and intra-predation loads, non-native flora and fauna, recreational activities, and trout management effects. The parameters that are measured and the methods employed in their measurement are described below for both the aquatic and the terrestrial environment.
Aquatic Environment - Parameters monitored in the aquatic environment are those deemed important for the survival of aquatic species.
1. Water parameters - Water Resources Division of the U.S. Geological Survey collects data from water gaging stations and collects and analyzes water samples at various locations within Lake Powell, the main channel, and within major tributaries.
a. Tailwaters to Lee’s Ferry sites - Every hour remote monitoring stations measure and record flow rate, sediment load, turbidity, water temperature, specific conductance (total dissolved solids) in the main channel. These data are transmitted by telemetry to a permanent storage location every four hours. In addition, dissolved oxygen and pH are measured using hydrolab mini-sonde instruments every 5 minutes for a 48 hour period during various parts of the year. Theese remote measuring devices are located at river miles 0, -3, -6, -9, -11, and -16. Monthly water samples are also collected at these locations and are analyzed for nutrients [P, soluable reactive P, dissolved ammonia nitrogen, total Kjeldahl nitrogen, dissolved nitrate-nitrite nitrogen], major ions [Na, Ca, Mg, K, SO4, Cl, CO3, (CO3)2], chlorophyll, phytoplankton, and zooplankton.
b. Main channel and tributary sites - Remote water monitoring stations measure the same water quality parameters at the same frequency as that described for the tailwaters within several tributaries (Paria River, Shinumo Creek, Tapeats Creek, Spenser Creek, Havasu Creek, Kanab Creek, Bright Angel Creek, Little Colorado River, and Diamond Creek) and within the main channel above the Glen Canyon dam, at Lee’s Ferry, above the Little Colorado River confluence, near the Grand Canyon, above National Canyon, and above Diamond Creek.
c. Lake Powell - Full lake-wide surveys are conducted quarterly (approximately March, June, September, and December) to obtain profiles of water temperature, specific conductance, dissolved oxygen, pH, turbidity, Secchi transparency, and weather conditions at approximately twenty stations north of Glen Canyon dam between mile 2 to mile 263. Water samples are also collected at some of these sites and are analyzed for nutrients [P, soluable reactive P, dissolved ammonia nitrogen, total Kjeldahl nitrogen, dissolved nitrate-nitrite nitrogen], major ions [Na, Ca, Mg, K, SO4, Cl, CO3, (CO3)2], chlorophyll, phytoplankton, and zooplankton. In addition, water quality profiles (water temperature, specific conductance, dissolved oxygen, pH, and turbidity) and biological samples (chlorophyll, phytoplankton, and zooplankton) are obtained on a monthly basis at the twenty monitoring stations.
2. Aquatic foodbase - At 10 day intervals for 90 days each year, foodbase is determined within pool, cobble-riffle, shoreline, and nearshore environments. The foodbase surveys are performed at river miles -15.5 (Glen Canyon gage), 0 (Lee’s Ferry), 60 (60-Mile Rapid), 64 , 138 (138.5-Mile Rapid), and 205 (cobble bar). For the pools, foodbase is examined at five locations along each of three transects; each transect is about 30 m apart. The five sampling locations along the transects include the thalweg, <28m3/s, baseflow, lower varial, and upper varial zones. Cobble riffle sample collections occur within the deepest accessible zone, as well as the lower and upper varial zones. Population data are collected for five biotic classes: (1) C. glomerata, (2) Oscillatoria spp., (3) detritus (both autochthonous [algal/bryophyte/macrophyte fragments] and allochthonous [tributary upland and riparian vegetation flotsum]), (4) misc. algae and macrophytes, and (5) macroinvertebrates (including Copepoda, Cladocera, Ostracoda, planaria, hydra, Gammarus lacustris, chrinomid larvae, simuliid larvae, lumbriculids, tubificids, physids, trichopterans, terrestrial insects, and unidentifiable animals). Associated data are also collected, such as water temperature, dissolved oxygen, pH, specific conductance, substratum type, microhabitat conditions, total P and N, Secchi depth, water velocity, depth, site, and time of day. Shoreline habitats are sampled to determine (1) invertebrates in emergent vegetation, (2) fine sediment volume, and (3) tychoplankton. Nearshore habitats are surveyed to obtain (1) temperature profiles with reading every 5 cm within shoreline vegetation and 0.5 m from the shoreline and (2) surface ( 0.5 m depth) drift samples of coarse- (500 micron mesh) and fine-particulate (0.5 micron mesh) organic matter.
3. Marshes - The populations of cattail-reed-watersedge (Carex aquatilis)-rush “wet” marshes and horsetail “dry” marshes are measured at 11 sites, if they exist, during the Labor Day terrestrial flora survey. The sites are located at river mile 6, 43.5, 50.5, 55, 68.5, 71.5, 93, 123, 194, 209, and 243. Future monitoring will be system wide.
4. Fauna - Three times per year (near January, June, and September) the age class, recruitment, and population size of Humpback Chub, Razorback sucker (very rare), Flannelmouth sucker, Bluehead sucker, and Speckled dace are collected, along with food production and habitat quality (temperature, suspended sediment, nutrient supply). The surveys are performed at Fence Fault Springs (RM 30), Little Colorado Confluence (RM 67), Bright Angel Creek (RM 87.7), Shinumo Creek (RM 108.8), Middle Granite Gorge (RM 127), Havasu Creek (RM 156.9), and Pumpkin Springs (RM 212.9). In calendar year 2001, surveys will concentrate more on the Middle Granite Gorge, Havasu Creek, and Little Colorado Confluence to determine if mainstem Chub are relicts of a mainstem spawning population or originate from the Little Colorado Confluence population. The Rainbow Trout will be monitored between Glen Canyon dam to the Paria River confluence four times a year at 30 locations that sample different habitats. The exact locations are yet to be determined for calendar year 2001.
5. Backwater areas - The presence and number of backwaters within or near return-current channels, shoreline embayments, and tributary mouths are monitored during the above faunal surveys.
Terrestrial Environment - Parameters monitored in the terrestrial environment are those deemed important for bank stabilization, aquatic and terrestrial faunal habitats, tribal botanical resource, and recreation.
1. Avifauna - Populations of Bald Eagle, Peregrine Falcon, SW Willow Flycatcher, passerines, waterbirds, osprey, and belted kingfisher are measured in January, February, April, May, and June. There are a total of 110 patches throughout the entire system that are monitored, although the exact number of patches monitored during any given year is less than this and varies year to year. This effort will be expanded in calendar year 2001 to include insects, icthyofauna, and small mammals that may be related to avifaunal distribution and productivity.
2. Other fauna - Some effort is expended in monitoring the existence of the endangered Kanab Ambersnail and less so of the non-endangered, but rare Niobrara Ambersnail and Northern Leopard Frog. The Kanab Ambersnail prefers Monkey flower, Watercress, and poison ivy and therefore occurs mostly at Vaseys Paradise. The Niobara Ambersnail prefers Typha and other wetland vegetation and occurs near river lime -9 and at Indian Gardens; it is currently not being monitored. Flows greater than 20,000 cfs inundate their habitats; such flows occurred every year since 1998. The Northern Leopard Frog has been found at river mile -9 and in Spring Canyon at river mile 204.
3. Flora - Vegetation type, area, and height are mapped annually at 11 sites (river miles, 6, 43.5, 50.5, 55, 68.5, 71.5, 93, 123, 194, 209, and 243) using aerial photographs and field studies in September (Labor Day). This survey will be expanded in the future to include the entire river system and culturally significant plants (exotic/invasive species and ethnobotanical species); sampling will occur in April and May, in addition to the usual September survey. Vegetation of interest includes Acacia, Equisetum/sedge, Redbud (Cercis occidentalis), Tamarisk (Tamarix ramosissima), Arrowweed (Tessaria sericea), Bermuda and red brome monotypic grasses, Hackberry (Celtis reticulata), Cliffrose, Desert brome, Mesquite, Coyote willow (Salix exigua), Baccharis seepwillow (Baccharis emoryi and salicifolia). Currently, downslope growth rates of Equisetum, Juncus, and Phragmites are being measured along selected transects on monthly basis for 4 months. The monthly field studies also monitor growth or removal of exotic plants, such as Tamarisk ramosissima and Alhagi camelorum (camelthorn), and sample low-elevation areas for changes in seed abundance and type.
Ecological Models: Korman and Walters (1998) have developed an ecological model for the Grand Canyon river system that is designed for planning adaptive management experiments. The software system allows for the selection and variation of a wide range of environmental parameters associated with the river system; the algorithm operates under these conditions to produce (predict ) changes in the environment. The model considers physical, aquatic, animal, and recreational/socio-economic factors using tabulated data and numerical codes to simulate the various processes.
Remote Sensing Recommendations: Aquatic Environment:
1. Water parameters - It is not possible to obtain from currently available remote sensors even a qualitative estimate of individual elements or compounds that are measured by water sample analyses, which was also stated in a more limited sense by the remote-sensing PEP (Berlin et al., 1998). A remote sensing device was developed in the early 1980's (Fraunhofer Line Discriminator) that detected ppm levels of phosphorous, but that development was terminated by NASA and the instrument did not reach operational status. NASA recently convened a panel of remote-sensing experts to address this very issue and the panel produced a set of recommendations for future sensor development and deployment that was as extensive as the panel membership. It will be many years before any of these recommendations reach flight status. Water parameters that can currently be measured using remotely sensed image data include sediment load (as total suspended sediment), turbidity, chlorophyll a and b, total chlorophyll, and possibly total dissolved solids (specific conductance). There have been many studies conducted within the last decade to develop remote sensing algorithms to measure these parameters in water (e.g., Goodin et al., 1993; McFeeters, 1996; George, 1997; Sathyendranath et al., 1997; Fraser, 1998a, b; Tassan, 1998). Of the algorithms that could duplicate in situ measurements with high accuracy, some algorithms were linear, while others required exponential relations. Most of the studies determined that multiple wavelength bands are necessary to obtain accurate estimates; this is consistent with a theoretical study that found that accurate estimates of phytoplankton, suspended mineral, and dissolved organic carbon contents could not be obtained using just one or even two wavelength bands (Pozdnyakov et al., 1998). A review of all of these results does not indicate a consistent (dominant) set of wavelength bands that best duplicate in situ measurements, although the bands that proved most useful were between 0.420 μm and 0.710 μm. [The wavelengths that were found useful for estimating water parameters consist of the following: 0.429 μm; 0.440 μm; 0.485 μm; 0.486 μm/0.570 μm ratio; 0.570 μm; 0.600 μm-0.520 μm difference; (0.600 μm-0.520 μm)/(0.570 μm-0.450 μm) ratio; 0.660 μm; 0.695 μm; 7 bands between 0.420 μm and 0.710 μm; 0.840 μm; and 23 bands between 0.72 and 0.95.] It is apparent from the past studies that the algorithms to estimate water parameters need to be derived for a particular water body and probably require continuous calibration with in situ measurements. This last statement (limitation) might suggest to water resources personnel that remote sensing cannot benefit their monitoring protocols because it cannot replace their in situ measurements, but the real strength in using remotely-sensed image data is, in most cases, not the elimination of field verification, but it is the extrapolation of site-specific information to wide areas at a significant savings of time and expense. The current research by Pat Chavez is designed to develop such algorithms for a few water parameters by monitoring spectral radiance of the water at specific water gaging stations (Chavez et al, 1999). This effort should be expanded to include (1) additional in situ radiometer sites in order to consider different aqueous conditions, including Lake Powell, (2) the use of more sophisticated radiometers with more bands in order to determine the best set of bands that provide the most accurate algorithms, (3) the consideration of more water parameters than under investigation, which are currently limited to the physical resource protocols, and (4) the use of airborne multispectral data to test the accuracy of the algorithms by extrapolation to nearby sites that are sampled by water resources personnel. A remote-sensing, image-analysis approach to monitoring particular water parameters is actually better suited to standing water bodies; George (1997) developed algorithms for lakes that can predict chlorophyll concentrations within 3 μg/l. Thus, this approach should seriously be considered for expanded areal mapping certain water parameters (chlorophyll, turbidity, suspended load) in Lake Powell, which was also suggested by the remote-sensing PEP (Berlin et al., 1998) and demonstrated by Chavez et al. (1997). However, it is not possible to derive the depth profiles of these parameters using remotely sensed image data. Even though the use of remotely sensed image data analyses for water parameters will probably increase somewhat the cost for such protocol monitoring, it will greatly increase the field area that is monitored, and may in the future reduce monitoring costs by targeting specific areas of change that should be sampled instead of continually monitoring the same sites each month or year.
It is also possible to map surface radiant temperature of water using airborne thermal-infrared sensors to a sensitivity of 0.1 degrees Celsius. This is most appropriate to non-turbid riverine and lacustrine environments, where it can provide near-surface water temperature maps that can either augment or possibly replace current lateral transects. However, thermal infrared sensors detect the near-surface temperature and thus will not provide thermal gradient information. Thermal-infrared data were collected in July, 2000 between river miles 30 and 74, coincident with a field survey in the area. These data will be examined during calendar year 2001 to determine the correspondence of the water temperatures measured by field profiles and by the airborne sensor. A recent study used in situ temperature and turbidity measurements in 3D hydrodynamic models to model effluent moving from Poplar Creek into the Clinch River in Tennessee (Garrett et al., 2000). The spatial distribution model results agreed qualitatively with temperature and turbidity maps obtained from an airborne multispectral scanner system; additional ground-truth data are being collected for a more quantitative assessment.
The water quality measurements are conducted at very short time intervals ranging from every few minutes to every 3 months. Augmenting these analyses at the minute interval is currently beyond spaceborne orbital capabilities (generally every 16 days) and would be extremely expensive using contractor airborne systems. However, bi-weekly or monthly data could be obtained from spaceborne multispectral sensors that have 4-m spatial resolution, which resolution would be more appropriate for Lake Powell. The usefulness of such data would depend on the wavelengths that were found to reliably estimate particular water parameters. Alternatively, GCMRC could purchase a multispectral sensor with wavelength capability between 0.4 and 0.9 and locally contract for periodic flights. This would provide flexibility in selection of wavelengths to be acquired and the spatial resolution and locations of the flights. In order to make such a sensor operational would require establishing a mount for the aircraft, a recording mechanism for the GPS and IMU, and a series of image-processing steps in transforming the point-perspective image data to a georectified or an orthorectified format.
2. Aquatic foodbase - Remotely sensed image data cannot however directly detect the small living organisms that are monitored in the foodbase surveys. In addition, the depth of penetration of remotely sensed data will depend on the clarity (turbidity) of the water and may not be able to reach the deepest parts of certain cobble riffles that are sampled by the foodbase surveys if the water is too turbid. However, some of the parameters that are measured related to aquatic foodbase can be approached using remotely sensed image data. These parameters include algae, vegetation flotsum, plankton, organic matter, surface drift, total dissolved solids (specific conductance), lateral water temperature profiles, depth, and substrate type. Published approaches to monitoring these parameters were discussed above for water quality and in the physical resource recommendations for the aquatic environment. In addition, Alberotanza et al. (1999) determined the best wavelength regions to detect specific characteristics of the aquatic environment. They found that the 0.430-0.510 μm wavelength region monitors absorptions due to organic matter, chlorophyll a, Beta carotene, and zeaxanthine; the 0.600 μm wavelength monitors the reflectance due to algal pigment; the 0.67-0.69 μm wavelength region monitors absorptions due to chlorophyll a; and the 0.700-0.710 μm wavelength region monitors reflectance due to absorptions on the sides of this region from algae, water, and chlorophyll a fluorescence. The remote-sensing PEP (Berlin et al., 1998) recommended that underwater videography be explored mapping algal mat and suspended algae. They also suggested that airborne multispectral image data be explored for this purpose. The airborne imaging approach is a better recourse because visibility will limit underwater videography in the same way that it limits airborne image data and because the exact area covered by videography is uncertain, variable, and difficult to record. Even though the scientists currently involved with monitoring foodbase are skeptical about a remote sensing approach to part of their protocols, a remote sensing approach should be explored within the aquatic remote sensing research that is being conducted by Pat Chavez. As in the case of the water quality surveys, the foodbase surveys are conducted at numerous time intervals during the year, but the issue regarding frequency of measurement versus cost for remotely sensed data could be overcome by using a GCMRC sensor, if one is procured for water quality and other high-frequency monitoring needs within GCMRC.
3. Marshes and Backwater Areas - The presence of both “wet” and “dry” marshes, as well as backwater areas, can be approached using remotely sensed radar or multispectral optical image data. Microwave (radar) energy is very sensitive to the moisture content of a surface material with increasing moisture resulting in a decreasing radar return signal. Rio and Lozano-Garcia (2000) used very simple image=processing filtering techniques on single co-polarized (Chh) radar images to map marshes in Laguna Madre wetlands of Mexico at an accuracy of 95%. The least expensive, “high-resolution” (10 m) radar image data is derived from spaceborne instruments ($4000 per 50 km x 50 km area), which would require $40,000 to cover the entire ecosystem once. The fact that the marshes and backwater areas can be relatively small in areal extent and that the marshes can be totally covered with vegetation, suggests that a radar approach is not appropriate because of its rather low resolution (10 m) and because of the fact that the radar signal will not penetrate the dense stands of phragmites. A more reasonable approach that can detect either moisture or a collection of particular vegetation species is the use of multispectral optical image data. Optical data are sensitive to variations in soil moisture (especially standing water) and can distinguish certain vegetation species (or collections) by their characteristics reflectance spectra. The detection and discrimination of the marsh and backwater habitats would be straightforward if the spectral reflectance of their vegetation collections (or lack thereof) was appreciably different. Jensen et al. (1993) could accurately map the annual distributions of cattails and water lilies using three-band SPOT image data, but only using seasonal image data in which the water lilies were senescent during part of the year. Such a scheme may not be necessary for the two types of marshes (one marsh containing mostly cattails, reeds, and watersedge and the other containing mostly horsetails) if wavelength bands could be found that optimize their discrimination. Ground reflectance spectra of these vegetation types have been collected during calendar year 2000 and will be examined to determine if identification of these two marshes can be accomplished. There are indications in the published literature that remote-sensing data can map these two environments. Welch et al. (1988) used 1:10,000- and 1:24,000-scale CIR stereo imagery to map emergent, submergent, floating, and mixed aquatic vegetation. Although they could not identify submergent plant species, they were able to identify the other plant collections based on height, color, and texture. Thomson et al. (1998) could distinguish 10 riparian classes (wet sand, diatoms and sand, wet silt/mud, algae and silt/mud, and six types of salt-marsh vegetation) using six wavelength bands (0.535-0.545 μm, 0.647-0.657 μm, 0.665-0.675 μm, 0.680-0.685 μm, 0.705-0.715 μm, and 0.870-0.890 μm) with a single wavelength band (0.647-0.657 μm) accurately separating all vegetated areas from non-vegetated areas. Marsh surveys are to become system wide in the future; a remote-sensing approach could satisfy that requirement and at a very low cost; an algorithm to detect a set of textural and spectral characteristics should be easy to develop assuming the remote-sensing data are calibrated throughout the Grand Canyon. Timing may be an issue because the optimum period for airborne acquisition is around the Summer solstice, but the marsh surveys are generally performed around Labor Day when low solar elevations cause shadowing.
4. Fauna - The direct detection of any fauna, especially in water, is extremely difficult and rarely approached using remote-sensing data due to the fact that their detection is based solely on recognition of their shape and size. Roberts and Anderson (1999) were able to visually detect fish populations using high-resolution photography (used for resolution) and multispectral CCD image data (used for spectral content), but they were actually trying to detect the off-channel habitats of the fish. They concluded that, if aerial photography is to be used for habitat mapping, the data need to be calibrated, which has been recommended previously in this report for any application. The habitats of sight feeders (trout) differs from that of the smell feeders (chub) in terms of turbidity levels, but other characteristics include the presence cobbles, vegetation, depth, and foodbase. Remotely sensed data can detect these characteristics, if the water is not too turbid; a set of these characteristics derived from remotely sensed data can be examined in a GIS environment to pinpoint a coincidence of a combination of selected characteristics or of all characteristics. Aquatic fauna target sites are examined 3-4 times per year. Small-area airborne data acquisitions are expensive because of mobilization costs. It would be better to use a GCMRC-owned multi-band sensor (as mentioned above) for such small-area temporal studies.
1. Land-based fauna and avifauna - Although remote-sensing data cannot directly detect fauna, especially at the size of birds, snails, frogs, and insects, it can be used to identify and map faunal habitats if the habitats have distinct characteristics. Sogge et al. (1998) showed that bird abundance in the Canyon is best predicted by total patch size (total land and vegetated areas) and by tree area, new-high-water-zone area, tamarisk area, tree volume, and shrub volume. He also found that bird richness and the Shannon diversity index was best predicted by tree area and volume, tamarisk area and volume, and new-high-water-zone area and that mesquite and tamarisk was most correlated with abundance, richness and diversity. He found tree volume to be the best predictor, but he felt that the extra field work necessary to obtain tree volume may not be worth the added expense because area predictors are almost as good as tree volume. Likewise, the Kanab Ambersnail prefers Monkey flower, Watercress, and poison ivy, whereas the Niobara Ambersnail prefers Typha and other wetland vegetation. It is likely that visible and near-infrared multispectral data can identify these vegetation collections and their physical attributes (such as height, volume, area), which would allow wide-area mapping of the habitats with only minimal human ground truthing. Such an approach was recently demonstrated for mapping the habitat of the Eurasian badger in the United Kingdom using a GIS model on data that reflect of the known characteristics of the badger habitat; the model produced an 80% accuracy in identifying badger habitats (Wright et al., 2000). A system-wide habitat mapping approach was also recommended by the remote-sensing PEP (Berlin et al., 1998).
Vegetation height or volume can also be approached without field measurement using optical or radar image data or LIDAR data. Using optical image data with known solar-elevation and solar-azimuth angles, a computer program can search for shadows cast by vegetation stands and automatically estimate their heights; the spectral signatures of the trees would allow automated mapping of tree area, thus providing tree volume. Such an analysis requires calibrated, orthorectifed image data; its operation beyond algorithm development would be essentially no cost. This approach differs from attempts to derive leaf area index (LAI, which is vegetation mass per unit area) using optical data; that approach has produced inconsistent results. Hurcom and Harrison (1998) found good correlation between NDVI and surface leaf area, but no correlation with LAI. Todd et al. (1998) found that the green vegetation index (where a green wavelength band replaces the red wavelength band in the NDVI), the NDVI, the wetness index (from a tasselled-cap analysis), and just a red wavelength band do not estimate biomass in grasslands very well. Blackburn (1999) found no mathematical relation between any spectral ratio index and LAI. Cusack et al. (1999) found an exponential relation between NDVI and biomass, but no relation to LAI. However, Blackburn and Steele (1999) found a good correlation between certain near-infrared spectral-reflectance ratios (0.836 μm/0.817 μm; 0.969 μm/0.931 μm; first derivative of 0.750 μm region; second derivative of 0.753 μm region) and unit leaf mass. More recently, Elmore et al. (2000) have determined that spectral mixing analysis using a pre-defined spectral library of all vegetation types can predict the percent area of live vegetation with an accuracy of 4 absolute percent, far superior to the results obtained using NDVI.
A more straightforward approach would be to acquire L-band radar imagery of particular areas of interest whose signal is scattered internally by tree trunks and branches, which causes the reflected signal to be depolarized relative to its incident polarization. This depolarization is called volume scattering and has been quantified by field analysis. The one pitfall of that approach is the level of interaction between the L-band signal and the vegetation decreases with decreasing branch/trunk density. Therefore, some vegetation might not be seen. If an autonomous approach is desired for habitat characterization, it would be better to use optical data, unless multiple wavelength (both C- and L-band) radar can be employed. Hill et al. (1999) found that L-band radar data are good for detecting short vegetation, but C-band radar data is better at detecting low herbaceous vegetation. Musick et al. (1998) found a linear relation between average stand density and Lhv backscatter coefficient. [Lhv is L-band data whose signal is sent with horizontal polarization and received at the antennae in only vertical polarization.] A third approach is to acquire LIDAR data with a large spot size (10 m for trees, so that it has a highly probability of hitting both vegetation and the ground surface) with an instrument (called the SLICER) that records the entire LIDAR return waveform from the surface. Means et al. (2000), using large LIDAR spot sizes, were able to accurately estimate tree height, basal area, and volume at one-half the cost of conventional methods. Also, very accurate estimates of vegetation volume, mean tree diameter at breast height, number of stems greater than 100 cm in diameter, basal area, biomass, and leaf area index have been obtained by LIDAR waveform analysis of large-diameter LIDAR signals (Lefsky et al., in press). The LIDAR surveys are less expensive and more accurate than the optical image surveys for measuring certain physical characteristics of vegetation stands, but a LIDAR approach should be fully evaluated for the vegetation stands with the Grand Canyon. With respect to the SLICER instrument, the broad spot size used in the LIDAR’s waveform analysis is not appropriate for detailed topographic mapping of the ground. Thus, a more universal LIDAR system should be explored to monitor both terrestrial topography and vegetation characteristics. Whether an optical or a LIDAR approach proves to be most viable, the selected system will offer wide-area, semi-automated analyses that will be more rapid and less expensive than human field surveys per unit area.
2. Flora - More advanced remote-sensing methods should be explored to replace the current methods for mapping terrestrial vegetation. At this time aerial photographic prints (just recently CIR) are used to manually trace the distribution of vegetation species, based on a visual interpretation of what CIR color and texture versus known vegetation occurrences. All derived polygons are then field checked. More appropriate wavelength bands should be considered for this analysis so that identification can be automated as much as possible using vegetation spectral signature and texture, both of which are easily assessed in commercial image-processing systems. In order to perform such an automated, or at least a aemi-automated, analysis requires calibrated image data, preferably orthorectified so that absolute areas and volumes can be automatically computed. The accuracy of an autonomous approach will be greater, the greater the spectral distinction between different vegetation species. Figure 5 shows spectra of some vegetation common to the southwestern U.S. Similar high spectral resolution data have been acquired for the different vegetation species that occur specifically within the Grand Canyon during calendar year 2000. These spectral data will soon be evaluated to determine the degree (accuracy) to which spectral mapping for vegetation can be autonomous just using spectral data. Of course, texture also needs to be considered in that analysis.
The presence and concentrations of several constituents control the spectral reflectance signature of vegetation: water, chlorophyll a (green; absorbs at 0.430 μm and 0.662 μm) and b (blue green; absorbs at 0.453 μm and 0.642 μm), assessory pigments (e.g., carbon-hydrogen rings of Beta carotene, lycopene; absorbs between 0.460 μm and 0.550 μm), nitrogen, lignin (cell wall polymer), cellulose (40-60% of cell walls), and open pore space. Thus, there has been voluminous research over the past three decades to find remote-sensing methods to accurately estimate these constituents in the hope that they will provide a quantitative tool for mapping vegetation type. Recent research has determined the following.
1. There is a strong linear correlation between chlorophyll (a+b) and the ratios 0.750 μm/0.700 μm and 0.750 μm/0.550 μm and the green vegetation index (Gitelson and Merzlyak (1997), the 0.700 μm and 0.735 μm reflectance and band ratio (Gitelson et al., 1999), the first derivative of the green vegetation index (Elvidge and Chen, 1995), the perpendicular vegetation index (Richardson and Wiegand, 1997), the first derivative of the 0.721 μm band (Blackburn et al., 1999), and the band ratios 0.836 μm/0.817 μm and 0.969 μm/0.931 μm, the first derivative of the 0.750 μm band, and the second derivative of the 0.753 μm band (Blackburn and Steele, 1999). Blackburn and Steele (1999) also found good correlations between their wavelength band ratios and derivatives and the carotenoid content; Blackburn et al. (1999) found that the derivative of 0.721 μm band correlated well with total chlorophyll, chlorophyll a, and chlorophyll b, but not so well with carotenoid content.
2. Penuelas et al. (1997) found some relation between the 0.900 μm/0.970 μm band ratio and plant water content, but it was very weak. Also, Hardy and Burgan (1999) did not find a good correlation between NDVI and plant moisture.
3. Kokaly and Clark (1999) found good correlations between spectral reflectance centered at 1.730 μm, 2.100 μm, and 2.300 μm with nitrogen and cellulose, but not with lignin.
4. Spectral reflectance from vegetation is effected by soil and litter cover, illumination angle, and shadows and methods have been devised to mitigate these effects (Lee and Marsh, 1995; Garcia-Haro et al., 1996; Todd and Hoffer, 1998; Blackburn, 1999; Pinder and McLeod, 1999; Yu et al., 1999; and Quackenbush et al., 2000).
5. Good vegetation classification results (>80% accuracy) can be obtained using remotely sensed data (Butt et al., 1998; Purevdorj et al., 1998; Coulter et al., 2000), use of seasonal data improves classification for deciduous vegetation (Grignetti et al., 1997; Mickelson et al., 1998), additional and narrower wavelength bands increase classification accuracy (Elvidge and Chen, 1995; May et al., 1997; Green et al., 1998), and airborne imagery provides better accuracy than spaceborne imagery due to higher resolution (Rowlinson et al., 1999; Zhu et al., 2000).
All of this research points to the distinct possibility that terrestrial vegetation mapping can become more automated, extensive, and less expensive using remotely sensed data and existing image-processing algorithms than it is at the present time. A crude demonstration of this potential is shown by taking a digital CIR image (Figure 6) and performing a simple NDVI analysis of the red and near-infrared bands to produce a map that displays all areas with vegetation (Figure 7), based on the dramatic increase in reflectance from the red wavelength to the near-infrared wavelength that is distinctly characteristic of most vegetation (Figure 5). This analysis took just 10 minutes. A regional analysis, even as simple as NDVI, requires calibrated image data, otherwise the analysis could take hours to complete to compensate for changes in vegetation signature between image frames.
Ecological Models Even though Korman and Walters did not intend on their model system to be used for detailed, quantitative predictions of the effects of particular unnatural and natural events within the Grand Canyon, it is very tempting to consider this software system as a ecological predictive tool. Even though the ultimate goal of the program was to develop management experiments, in the process of improving the model with additional, more regional or system-wide results that can be obtained by an expanded remote sensing program, the model may become the single most powerful tool for testing integrated cause/effect relations of dam operations.